Glossary

The document control glossary

36 definitions, and above all the distinctions that matter: OCR is not extraction, extraction is not control, document management is not compliance, a technical log is not an audit trail.

Affordability ratio

Ratio of rent including charges to a candidate's monthly net income, expressed as a percentage. Used as a solvency criterion in property management.

Not to be confused with

The result depends entirely on which net figure is used. Computing on net pay rather than taxable net, or on a month containing a bonus, shifts the ratio by several points.

AI Act

European regulation governing artificial-intelligence systems by risk level, requiring in particular transparency, traceability and effective human oversight for uses that significantly affect people.

See also : Human in the loop · Explainable decision · GDPR

AML (anti-money laundering)

Framework for combating money laundering and terrorist financing, of which KYC is the onboarding component.

See also : GDPR · Retention period

Document AI

Umbrella term for applying learned models to document understanding: classification, extraction, answering questions about a document. Often used as a product name by cloud providers.

Not to be confused with

"Document AI" describes a technology, not a guarantee of outcome. A model can produce a plausible-but-wrong value with the same confidence as a correct one — hence the need for a confidence threshold and human review.

Document classification

Automatic identification of an uploaded document's nature: payslip, tax notice, identity document. It determines which extraction and control rules apply.

Not to be confused with

A single PDF can concatenate several documents. Classifying the whole file from its first page is a frequent and costly mistake.

Document compliance

State of a file whose required documents are all present, valid and compliant with the applicable rules — internal or regulatory — at the date compliance is established.

Not to be confused with

Compliance is dated. A file compliant on signature day stops being so when a certificate expires: compliance is a state to maintain, not a milestone passed once.

See also : Document control · Audit trail

Document consistency

Property of a set of documents whose shared information agrees and whose internal calculations verify. Its absence is the first anomaly signal, whether fraudulent or accidental.

See also : Cross-document check · Document fraud

Document control

The set of checks applied to a document or a group of documents to establish whether they satisfy a set of rules: presence, validity, internal consistency, agreement with other documents. Document control produces a reasoned decision — compliant, non-compliant, needs review — and the evidence for it.

Not to be confused with

Not to be confused with extraction: extraction reads the document and produces data, control applies rules to that data and produces a decision. A system that extracts without deciding is not document control.

Document data extraction

Turning a document's content into structured, typed data: an amount becomes a number, a date becomes a date, a name becomes an identity field. Extraction relies on OCR for image documents but differs from it by qualifying values semantically.

Not to be confused with

Extraction produces data, not judgement. Correctly extracting an IBAN says nothing about its legitimacy.

Document fraud

Producing or altering a document to obtain an undue right, contract or payment: altered amount, wholly fabricated document, genuine document belonging to someone else.

Not to be confused with

Modern document fraud is rarely crude. A competent forger produces an arithmetically consistent document: it is cross-referencing several documents, not examining one, that catches them out.

Document workflow

Ordered sequence of steps applied to a file: collection, extraction, control, decision, possible escalation, archiving. The workflow makes processing reproducible from one file to the next.

See also : Document control · Escalation · Human in the loop

EDM (electronic document management)

System for storing, indexing, versioning and sharing documents. An EDM answers "where is the document and who may see it".

Not to be confused with

An EDM files; it does not verify. It knows a document exists and where it is; it has no idea whether it is expired, inconsistent or forged. The two systems are complementary, not competing.

See also : Document workflow · Document control

Escalation

Routing a file to human review when the rules cannot conclude: uncertain data, unforeseen case, detected anomaly whose cause remains ambiguous.

Not to be confused with

Escalation is not a system failure: it is its safety valve. An engine that never escalates is deciding cases it should not have decided.

Escalation rate

Share of files routed to human review. Complement of the automation rate, it measures the residual load on teams.

See also : Automation rate · Escalation

Explainable decision

Decision whose reasoning can be restated in terms the affected person understands: which rule applied, to which data, extracted from where in which document.

Not to be confused with

Showing a score is not explaining. "Risk score: 0.72" is not an explanation; "the insurance certificate expired on 14 March" is.

See also : Audit trail · AI Act · Human in the loop

False negative

File declared compliant when it is not: forged document accepted, inconsistency undetected, expired document approved.

Not to be confused with

A false negative only surfaces when it becomes expensive — an unpaid debt, an inspection, a dispute. It cannot be measured in the live flow: it must be sought by sampling.

See also : False positive · Document fraud · Automation rate

False positive

An anomaly flagged by a check on a file that is in fact compliant. Main cost: wasted human time and, if rejection is automatic, harm to a person whose file was valid.

Not to be confused with

A false positive is visible — someone complains. A false negative is not. Optimising only what surfaces leads to loosening rules until nothing is detected at all.

Example : Rejecting a 2011 paper ID card as expired when it benefits from the five-year extension.

GDPR

European regulation governing personal-data processing: defined purpose, minimisation of collected data, limited retention, rights of data subjects.

Not to be confused with

It creates a real tension with auditability: deleting a document on schedule is mandatory, so is proving it was checked. The answer lies in retaining the fingerprint and outcome, not the document.

See also : Data minimisation · Proof retention · AI Act

Human in the loop

Arrangement in which a human operator retains the effective ability to check, correct or overturn an automatically produced decision before it takes effect.

Not to be confused with

Purely formal oversight does not count. An operator approving a hundred files a minute with no real power to challenge is not human oversight, whatever the interface.

See also : AI Act · Escalation · Explainable decision · False positive

IDP (Intelligent Document Processing)

Software category combining classification, OCR, extraction and sometimes validation, to process unstructured documents at scale. IDP industrialises document reading.

Not to be confused with

IDP is measured by how well it reads; document control by how correctly it decides. An IDP platform can extract data flawlessly that, cross-referenced, would reveal an inconsistency — without ever performing that cross-reference.

KYC (know your customer)

Set of obligations requiring verification of a customer's identity and source of funds before entering a business relationship, then keeping that knowledge up to date.

Logging

Chronological recording of a system's technical events: calls, errors, connections. Intended for operations and diagnostics.

Not to be confused with

Logging answers "what happened in the system". The audit trail answers "why was this decision made on this file". Conflating them means handing logs to an auditor.